1 of 20

Modeling of Clouds from a Single Photograph

2 of 20

INTRODUCTION

  • Two major approaches to modeling clouds:

​

  • - Procedural approach
    • Low computational cost
    • Parameters have to be determined manually by trial and error

​

  • - Physically based approach
    • High cost
    • Only simulating the physical process

3 of 20

INTRODUCTION

GOAL：use the photograph as a guide to synthesize that look similar to those in the photograph. (Not reproduce exactly the same clouds.)

​

Three types of cloud：

Cirrus(卷雲)、Altocumulus(高積雲)、Cumulus(積雲)

​

  • Cirrus is generally thin and self-shadows are seldom observed. Therefore, cirrus is modeled as a two-dimensional texture.

​

  • Altocumulus and cumulus possess volumetric features and three-dimensional density distributions must be generated.

​

4 of 20

RELATED WORK

DOBASHI Y., NISHITA T., YAMASHITA H., OKITA T.: Using metaballs to modeling and animate clouds from satellite images. The Visual Computer 15, 9 (1998), 471–482.

5 of 20

OVERVIEW

6 of 20

CALCULATION OF CLOUD IMAGE

  • The first process is to create the sky image by estimating sky colors behind the clouds in the input image.

​

  • Then, the intensity and the opacity of the clouds are calculated by comparing the input photograph with the sky image.

7 of 20

CALCULATION OF CLOUD IMAGE

  • Calculation of sky image
  • To create the sky image, each pixel in the input image is roughly classified into either a cloud pixel or a sky pixel, and the cloud pixels are removed

​

  • The sky image is created by extending the colors of the surrounding sky pixels into the removed cloud areas

​

8 of 20

CALCULATION OF CLOUD IMAGE

  •  

9 of 20

CALCULATION OF CLOUD IMAGE

  •  

10 of 20

CALCULATION OF CLOUD IMAGE

  • Calculation of Cloud Intensity and Opacity
  • The intensity of light for k(> 1)th order scattering can be represented by the following recursive equations :

​

​

​

​

  • ω’ = direction from y to x
  • G = the form factor calculated by the geometric

relation between x and y

  • θk = angle between ω and ω’
  • Vc = all the points within the cloud volume

​

11 of 20

CALCULATION OF CLOUD IMAGE

  • Calculation of Cloud Intensity and Opacity

​

12 of 20

CALCULATION OF CLOUD IMAGE

  • Calculation of Cloud Intensity and Opacity
  • The intensity of light reaching the viewpoint is obtained by accumulating Im along the viewing ray
  • The light of the sky behind the clouds also reaches the viewpoint after being attenuated by the cloud particles

​

​

​

  • T = the thickness of the clouds

​

13 of 20

CALCULATION OF CLOUD IMAGE

Isun(λ) = isun csun(λ)

Combine Eq. (3) and Eq. (4)

​

​

​

​

​

​

​

​

​

α = opacity

β = intensity

14 of 20

MODELING OF CLOUDS

  • Cirrus(卷雲)

​

User assumed

15 of 20

MODELING OF CLOUDS

  • Altocumulus(高積雲)

​

​

​

​

  • Typically consist of many small cloud cells

​

  • The basic idea is to generate a set of metaballs to approximate the cloud image

​

​

16 of 20

MODELING OF CLOUDS

  • Altocumulus(高積雲)
  • (b) a set of binary images Ik,B(k = 1,2,···,n) are

created by using a set of thresholds for α

​

  • (c) a distance transform is applied to each of

Ik,B and a distance image, Ik,D is created

​

  • (d) Each pixel in Ik,D stores the distance to the

nearest black pixel. Metaballs are then

generated at the locations of all white pixels of Ik,B

17 of 20

MODELING OF CLOUDS

  • Altocumulus(高積雲)
  • (e) In order to make the synthetic clouds look like those in the input photograph, the center densities of the metaballs are determined so that the cumulative density at each pixel becomes the same as the cloud intensity β(p).

​

​

​

  • (f) With a similar method to that used for cirrus clouds, the cloud plane is specified by the user, and the center positions of the metaballs are projected onto the plane

​

​

18 of 20

MODELING OF CLOUDS

  • Cumulus(積雲)
  • (a)

Cloud image is converted into binary image.

Distance transform is applied to the binary image and medial axes are extracted.

The colors of the medial axes correspond to the distances: the blue color corresponds to zero and the red color to the longest distance.

  • (b) (c)

Minimize the energy function

19 of 20

MODELING OF CLOUDS

  • Cumulus(積雲)
  • (d) The density distribution inside the surface shape is generated by invoking the Perlin Noise.
  • First, a binary volume is created by subdividing the bounding box of the surface shape into a regular grid.

​

  • Then, the three-dimensional version of the distance transform is applied to the binary volume.

​

  • Next, the distance is normalized so that the maximum distance becomes 1.

​

  • Use the product of the Perlin noise and the normalized distance as the density at each grid point.

20 of 20

RESULTS